Using pairs of genes where one is present at the same location as a trans-QTL generated by the other, a directed interaction network can be created. The target of each interaction pair is the gene, the expression pattern of whose transcript exhibits a trans-QTL. The source node of each interaction is a gene whose genomic location lies within the QTL interval. Figure3.4illustrates this concept. The resulting network consised of 499358 edges connecting 29856 nodes—2457 of which were target nodes and 29791 source nodes. The target nodes represent all genes exhibiting a QTL—1588 cis-acting + 2120 trans-acting, with some overlap where a gene had both a cis- and a trans-QTL. The high number of source nodes was a result of the many genes in each polymorphic interval, each of which gave rise in the network to an edge pointing toward the target. This was visible as a star motif in the network graph (not shown) of which the target trans-regulated gene was at the centre with a high indegree. This was typical for all trans- regulated genes and resulted from the rather large linkage blocks inherent in the BXD cross and the broad QTLs these lead to. The assumption is that only one of the genes in the QTL interval is the ‘real’ causal gene. Ultimately, all variation in the BXD panel must come from genomic polymorphism, so true candidates for the source gene must contain a SNP (or other sequence variant) and will also likely, although not necessarily, exhibit a cis-QTL. Another motif apparent in the trans-QTL network was a few-to-many star, where a single locus exerts wide-ranging pleiotropic effects on the transcription of many downstream factors. This pattern suggests the action of a ‘master’ regulatory element and is of interest in the search for potential therapeutic targets. This effect is more clearly observed as the trans-band pattern discussed above.
Discussion
The dataset described here presents a rich mine of QTLs—each representing a hypothesis linking transcript expression to a sequence variant in the genome. The majority of these are cis-QTLs implying that variation in the gene itself is responsible for expression variation. Many, however, are trans-QTLs which are particularly interesting because they suggest a link between two genes. This is an important step in reconstructing the molecular pathways leading to the phenotype being studied. Here, the gene-gene links have been formalised as a directed interaction network similarly to a previous report (Li et al.,2005). This network is highly redundant, as every gene within the QTL interval is used as an interaction source node. Due to the very large size of
44 Interactions Between Phenotype and Genotype
Figure 3.4: A graphical explanation of trans-QTL edges. The upper panel shows two QTL traces; the source gene exhibiting a cis-QTL at the same location as the gene (depicted by an orange arrowhead), and the target gene with a trans-QTL at the same locus as the source gene, but at a different location to its own gene position. The polymorphism giving rise to the QTLs is connected, not only to the source gene in cis, but also to the target gene in trans. Furthermore, the nature of the interaction makes this a causal link. The lower panel depicts a formalised representation of this relationship with the two genes as network nodes and two directed edges; a self edge at the source indicating the cis-QTL, and a gene-gene interaction, derived from the
trans-QTL relationship, linking the source and target genes.
the linkage blocks in the BXD population, however, the QTL intervals contain an average of 217 known genes. With the assumption that only one polymorphic gene is responsible for each unique QTL, this would mean that, on average, only 1 in 217 (less than 0.5 %) of the predicted edges in this network will be true positives. Nevertheless, when this network is only employed as a layer in a multigraph, the direction information it can lend to otherwise validated edges is an important addition to the gene-gene interaction landscape. While a number of other methods have been proposed to narrow down the list of candidates in the QTL interval (Li et al.,2005; Bing and Hoeschele,2005;Kulp and Jagalur,2006), these are, in essence, the addition of a single additional layer—for example expression correlation data. I had experimented with limiting the source candidates to only those genes which also displayed a cis-QTL with the reasoning that transduction of the polymorphism from the source to target gene might be expected to follow via changes in the expression of the polymorphic gene itself. While I do not agree with the decision
Interactions Between Phenotype and Genotype 45
of Kulp and Jagalur (2006) to expressly discard these candidates, this criterion did seem to be overly restrictive, so here all genes remain as candidate source nodes. While most of these source genes are associated with only one edge, a few are hubs with very high outdegree. These are the trans-bands referred to in figure3.2. Although trans-bands have been noted in the literature several times (Chesler et al., 2005; Mozhui et al., 2008; Loguercio et al., 2010; Overall et al., 2009), it is still not clear what the real cause of these is. The early assumption that trans-QTLs would largely be found to harbour transcription factors has not stood up to testing (Yvert et al., 2003;Kulp and Jagalur,2006) and the evidence suggests that the types of genes involved might be more diverse (Mozhui et al.,2008;Loguercio et al.,2010). The trans-band Trans5a, described above is currently the subject of a follow-up study in our laboratory and it will be interesting to see which gene is ultimately responsible for this hippocampus-specific source of pleiotropic regulation. All too often, however, the trail does not lead back to a single genomic locus, and this is particularly the case with phenotypes regulated by many genes together. The QTL approach was originally developed to identify genomic loci associated with complex phenotypes (Sax, 1923). Some phenotypes, however, are so complex that single loci can no longer be identified, and this is particularly the case for physiological traits which include many components at many levels (molecular, cellular, anatomical and even behavioural). Our own laboratory has struggled with unearthing significant genomic loci affecting adult neurogenesis traits and has looked to transcript expression for help (Kempermann et al.,2006). Because mRNA production is much more closely linked to gene sequence than are organism-level traits, it will more likely reflect segregating polymorphism. Treating transcript expression as a phenotype for QTL mapping (Jansen and Nap, 2001) provides large numbers of often very strong associations to genomic loci. Correlation of physiological traits to gene expression traits (Chesler et al., 2005; Overall
et al., 2009) thus offers a step in the right direction towards genomic linkage. The next step can be taken by identifying loci associated with the correlated transcripts to build a hypothetical causal path from polymorphism to observed phenotypic outcome. This methodology, which I have termed proxy -QTL mapping, and the intuitive graphical overview (as in figure 3.3) will hopefully be helpful in the future mapping of very complex phenotypes, and is already being applied in our laboratory with promising results.
4. Strain-Dependent Effects of Environmental
Influences on Neurogenesis
The preceding two analyses have made use of baseline measurements of transcript expression—the animals were sampled from a home-cage environment in the absence of any environmental perturbation. It is clear, however, that in addition to environmental in- fluence and the influence of genetic background under baseline conditions, a combinatorial response is also likely. The experiment described in this chapter addresses how genetic background influences the response to environmental stimuli that are known to affect the rate of neurogenesis. Mice from the two inbred strains, C57BL/6 and DBA/2, were housed under one of three conditions; running (RUN), enrichment (ENR) or in standard cages (STD). In a series of experiments, hippocampal gene expression was assessed using mi- croarrays and histology was used to estimate progenitor cell division. While DBA/2 mice had generally lower levels of proliferation in the hippocampus, the increase associated with voluntary activity that is well-known in C57BL/6 animals was not seen in DBA/2 mice. In addition, the DBA/2 animals housed in an enriched environment showed higher relative levels of expression of transcripts associated with synaptic plasticity.
Parts of the work described in this chapter have been published as Overall et al. (2013). Delayed and transient increase of adult hippocampal neurogenesis by physical exercise in DBA/2 mice. PLoS ONE (8)12 e83797.
Introduction
When mice have access to a running wheel in a standard cage, they will use it to run throughout the hours of darkness almost continuously—often clocking up extraordinary distances. It has been shown that such activity is a potent inducer of proliferation in the hippocampus (van Praag
et al., 1999a,b). Since this seminal work, the effect of RUN-induced neurogenesis (RING) has been replicated many times and has been shown to have an effect even after just one night of activity (Steiner et al., 2008).
48 Strain-Dependent Effects of Environment
A perhaps even more striking finding was the discovery that neurogenesis can be influenced by the complexity of the environment (Kempermann et al., 1997b). Firmly couched in urban myth as the “use it or lose it” catchphrase, it transpires that new neurons, at least in the hippocampus, may indeed be generated just when and where they are needed. Environmental enrichment can be provided in the laboratory setting by housing the animals in a larger cage with toys and a tunnel system to explore. Because the function of the hippocampus is thought to be especially important for spatial learning, the opportunity to investigate a more complex environment is expected to offer a good model of hippocampal stimulation. It is particularly intriguing that physical activity and environmental enrichment appear to affect different stages of neural precursor cell development (van Praag et al.,1999b;Kronenberg et al.,2003). This finding provides a tool to help investigate the genetic pathways regulating neural precursor proliferation and maturation.
Almost all previous work, however, has been done using the standard laboratory strain C57BL/6. This is despite an established strong effect of genetic background on precursor prolif- eration and new neuron production (Kempermann et al.,1997a;Kempermann and Gage,2002a; Kempermann et al., 2006;Clark et al.,2011). Inspired by the discovery of clear differences in genetic control of neurogenesis between C57BL/6 and DBA/2 (Kempermann and Gage,2002a; Kempermann et al., 2006), and to complement our extensive use of these strains, I investi- gated the effects of running and enriched environment in both C57BL/6 and DBA/2 in parallel. The experiments employed a factorial design, with animals from each of the two different ge- netic backgrounds housed in either standard cages (STD), standard cages with a running wheel (RUN), or in larger custom-built cages containing various objects and a labyrinth of tubing to allow scope for exploration (ENR). It is known from previous work with the strain C57BL/6 that wheel running and enriched environment stimulate the proliferation of the precursor pool in the dentate gyrus of the hippocampus (Kempermann et al.,1997b;van Praag et al.,1999b; Steiner et al.,2008), although the effect of ENR is rather at the level of survival of new neurons than proliferation (Kronenberg et al.,2003). The acute effect of housing environment on adult neurogenesis in different strains is, however, largely still unknown (but seeKempermann et al., 1998a). I discovered that a short (4 d) running stimulus, while sufficient to induce increased proliferation in C57BL/6, does not do so in DBA/2 mice. Using expression microarrays, I could also show that this strain-dependent difference in response to environment is accompanied by transcription changes in the hippocampus.